Evaluation of State-of-the-Art Classifiers: A Comparative Study
Researchpedia Journal of Computing, Volume 1, Issue 1, Article 3, Pages 22-29, June 2020
Abdulwahab Ali Almazroi1, Osama Abdo Mohamed 1,2, Azra Shamim1, Muhammad Ahsan Qureshi1,*
1College of Computing and Information Technology, University of Jeddah, Khulais
2 Faculty of Science, Math. Department, Zagazig University, Egypt
Corresponding author: Muhammad Ahsan Qureshi (e-mail: maqureshi@uj.edu.sa).
ABSTRACT Breast cancer is considered the most common deadly disease among women around the globe. The accurate detection of breast cancer at the early stages is a prime concern now a day to save many lives. In literature, different computer-aided detection systems are employed for breast cancer detection. These systems detect breast by classification cells as normal, benign, and malignant using data classifiers. Therefore, the current study evaluates the performance of state-of-the-art classifiers for breast cancer detection. Specifically, seven widely used classifiers: Stochastic Gradient Descent (SGD), Support Vector Machine (SVM), Nu Support Vector Machine (Nu SVM), Linear Support Vector machine, KNeighbors, Naive Bayes, Random forest, and ExtraTrees are compared based on their accuracy and execution time for breast tumor identification. The algorithms are evaluated on a dataset comprised of 569 cases (357 malignant, 212 benign). The experimental results show ExtraTrees classifiers outclassed other classifiers in terms of accuracy by achieving 97% accuracy. On the other hand, the Naive Bayes classifier exhibited the best performance in execution time. The least accuracy is accomplished by the SGD classifier. The highest execution is required by Linear SVC. This study guides developer of breast cancer detection systems about the selection of appropriate algorithm for their system.
Keywords Classifiers; ExtraTrees; SGD; SVC
